-
Notifications
You must be signed in to change notification settings - Fork 0
/
TFmodeling.py
76 lines (63 loc) · 2.19 KB
/
TFmodeling.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
import tensorflow as tf
from tensorflow.keras.layers import Dense, Dropout
from transformers import TFCamembertModel
INPUT_IDS_KEY = 'input_ids'
ATTENTION_MASK_KEY = 'attention_mask'
OUTPUT_KEY = 'classifier'
class CamemBertMultilabelClassification(tf.keras.Model):
def __init__(self,
nb_class,
name="CamemBertMultilabelClassification",
**kwargs):
super(CamemBertMultilabelClassification, self).__init__(name=name, **kwargs)
self.nb_class = nb_class
self.l1 = TFCamembertModel.from_pretrained("camembert-base")
self.pre_classifier = Dense(768)
self.dropout = Dropout(0.1)
self.classifier = Dense(self.nb_class)
def call(self, inputs):
output_1 = self.l1(input_ids=inputs[0], attention_mask=inputs[1])
hidden_state = output_1[0]
pooler = hidden_state[:, 0]
pooler = self.pre_classifier(pooler)
pooler = tf.keras.activations.tanh(pooler)
pooler = self.dropout(pooler)
output = self.classifier(pooler)
return output
def build_camembert_model(nb_class, seq_length, learning_rate):
"""
Build camemBert model for multilabel classification
:param nb_class: Number of class
:param seq_length: The sequence length
:param learning_rate: Learning rate for training
:return: The model
"""
# Input
input_ids = tf.keras.layers.Input(
shape=(seq_length,),
name=INPUT_IDS_KEY,
dtype=tf.int32
)
attention_mask = tf.keras.layers.Input(
shape=(seq_length,),
name=ATTENTION_MASK_KEY,
dtype=tf.int32
)
camemBert_model = CamemBertMultilabelClassification(nb_class=nb_class)
# Output
camemBert_output = camemBert_model(
inputs=[input_ids, attention_mask]
)
# Initialize the model
model = tf.keras.models.Model(
name="CamemBert",
inputs=[input_ids, attention_mask],
outputs=[camemBert_output]
)
model.compile(
optimizer=tf.keras.optimizers.Adam(learning_rate=learning_rate),
loss=tf.losses.BinaryCrossentropy(from_logits=True),
metrics='acc'
)
model.summary(line_length=130)
return model